Daydreaming of alternate universes & a tech marriage made in heaven: BeOS on Acorn

[liam_on_linux]

Acorn pulled out of making desktop computers in 1998, when it cancelled the Risc PC 2, the Acorn Phoebe.

The machine was complete, but the software wasn't. It was finished and released as RISC OS 4, an upgrade for existing Acorn machines, by RISC OS Ltd.

by that era, ARM had lost the desktop performance battle. If Acorn had switched to laptops by then, I think it could have remained competitive for some years longer -- 486-era PC laptops were pretty dreadful. But the Phoebe shows that what Acorn was actually trying to build was a next-generation powerful desktop workstation.

Tragically, I must concede that they were right to cancel it. If there had been a default version with 2 CPUs, upgradable to 4, and that were followed with 6- and 8-core models, they might have made it, but RISC OS couldn't do that, and Acorn didn't have the resources to rewrite RISC OS to do it. A dedicated Linux machine in 1998 would have been suicidal -- Linux didn't even have a FOSS desktop in those days. If you wanted a desktop Unix workstation, you still bought a Sun or the like.

(I wish I'd bought one of the ATX cases when they were on the market.)

Castle Technology continued with new desktop machines -- the Iyonix and others -- but they never made sense in bang-for-the-buck terms. I'd have loved a slim light RISC OS laptop, though.

An idea occurs I've never had before: 1998, when the Phoebe was cancelled, is the same year that Be ported BeOS to x86. A multi-processor Acorn machine running ARM BeOS would have been a stunning machine at the time -- what BeOS did on limited hardware, without acceleration, was superb. I'm surprised this has never struck me before.

Acorn hardware with Be software would have been a gorgeous combination, and would have offered a fairly compelling proposition for the time. As radical in terms of "power without the price" as the Atari ST was a decade before.

But looking at the bigger picture -- the rising ascendance of Windows NT, by that time a decent credible OS; Apple's provident acquisition of NeXT; and the fact that Linux was getting serious, with Kernel 2.0, proper SMP support, and KDE 1.0 coming out that year -- no, whereas there might have been a flurry of interest, it would probably have been dead within a few years.

So no, the Phoebe could not have competed. The 200MHz StrongARM wasn't massively quicker than the Intel Pentium 90, as the Pentium was a superscalar chip.

But it was cheap and it didn't put out a lot of heat.

Before Acorn launched the StrongARM upgrade for the RISC PC, another company, Simtec, offered a multiprocessing add-on, the Hydra. It allowed a Risc PC to have 4 additional ARM processor daughterboards in its second processor slot.

Symmetric multiprocessing -- SMP -- was niche at first but it was blatantly obvious for years that this was the coming thing. But in the late '90s, CPUs were becoming a significant part of system cost, and heat dissipation was becoming an issue. I saw machines failing when their cooling systems failed, or you even left the lid off.

There was a niche for Acorn to aim at, but it didn't, because it was committed to RISC OS. However, the foundations of that OS were not sound. Same as classic MacOS, same as Win9x.

Apple fooled around for far too long before acquiring NeXT and doing a remarkable big-bang approach rip-and-replace OS upgrade. I'm amazed it worked, but boy, did it ever work well for them. The dev tools, the design and leadership genius of Steve Jobs, and some great hardware were part of it too, of course.

Microsoft made the best of the OS/2 debacle and Portable OS/2 codebase, hired Dave Cutler, and in a magnificent achievement of avoiding the Osborne effect, managed to nurse the whole damned PC industry along on Win9x until NT was ready for prime-time.

Acorn didn't, stuck with its decrepit old last-minute hack of an OS, and then gave up. Its stub remnant of ARM did extraordinarily well, though.

Commodore and Atari didn't, and died, too. If Commodore had executed better on either its Taos/Intent or its QNX plans, it might have made it.

But they didn't, and we're perilously close to a monoculture now as a result.

I have an Amiga (and a Mac Mini G4 with MorphOS), I have an ST, I have a QL. I recently sold my BlackBerry, my only QNX device. I've played with them all, but I didn't know them well and I didn't love AmigaOS or GEM or QDOS or any of them.

The ones I used a lot and really did love were Psion's EPOC16 and EPOC32, classic MacOS and RISC OS 2. (By RISC OS 3 and later releases, I'd moved on.)

But while I loved the elegance of the GUIs of both RISC OS and MacOS 6/7/8/9, they were always terribly unstable and had deep problems caused by poor, or at least dated, design.

But there is precedent, remarkably, for making and selling SMP computers even if your OS doesn't support it. The Macs went SMP in '95 (!) with the Daystar Genesis MP and later Apple's own PowerMac 9600.

Classic MacOS didn't support the extra CPUs. Only PhotoShop did at first. That was enough to sell some kit, even at over $7,000.

I said back then that Acorn should do it. Of course, nobody listened.

Only Acorn could have sold multiprocessor machines which were not significantly more costly than uniprocessor ones, because in Acorn kit, the CPU was cheap and didn't require any cooling. These were plastic-packaged chips. Not even heatsinks. They could have shipped a dual-CPU machine for under £50 more than a uniprocessor one, I reckon. Nobody else could have touched that.

The big problem is that Acorn was committed to RISC OS by then, and RISC OS was and is resolutely a single-core, single-tasking OS with a cooperative-multitasking GUI on top -- a little like Windows 3 on MS-DOS. Acorn had already discontinued its Unix, RISC iX, in 1992.

Linux wasn't ready to be a mainstream desktop SMP OS yet at the end of the 1990s -- the first version to support SMP, kernel 2.0, came out in 1996 and I'm told it only usefully supported 2 processors. KDE 1.0 wouldn't ship until 1998.

So what OS would this hypothetical SMP Acorn workstation have run?

Well, it occurs to me now, that there was an ideal candidate. BeOS.

BeOS showed that if you took the design attitude of those 1980s GUI pioneers, and applied it to 1990s technology, it was viable and possible to create something really rather wonderful.

One of the bits of accepted doctrine that I really dislike is that the evils and the virtues of the 1980s greats -- such as MacOS and AmigaOS -- are inextricably linked.

By virtues, I mean things like:

* They were small and efficient.

* They were fast and responsive.

* They ran well on relatively limited hardware.

* They needed little storage and were even usable on floppy-only systems.

By evils, I mean things like:

* No multitasking (MacOS, DR GEM).

* Or only cooperative multitasking (Windows 3, RISC OS).

* Or proper pre-emptive multitasking, but without memory protection (AmigaOS).

* Poor networking.

These things aren't linked at all. They're artefacts of the design processes.

MacOS wasn't designed to be multitasking -- the Lisa was, but this was something removed from the plan to keep costs down.

RISC OS and Amiga OS were both last-minute rescue efforts, when the original all-singing all-dancing fully-preemptive OS development projects (ARX, CAOS) overran so badly that they had to be abandoned.

GEM was limited, yes, but it was an ingenious combination of off-the-shelf bits of OS combined with a very cleverly cut-price hardware design to deliver "power without the price" -- most of the attention-grabbing bits of the Mac (sharp monochrome WIMP GUI, 68K chip, 3.5" drives) combined with the colourful graphics and stereo sound of the Amiga, all for a budget price. It wasn't some amazing ground-up design like the Mac, the Amiga or the ST. It was a 68K plus some PC components and an OS put together from stuff on the shelves of Digital Research, the company that lost out on the IBM PC deal.

But BeOS showed that if you tried to design a clean, ground-up PC OS with the sine-qua-non requirements of the next generation of micros from late 1990s: preemptive multitasking, SMP, a journalling filesystem, TCP/IP networking, rich multimedia, etc. -- you could actually deliver.

BeOS had the virtues of the Amiga -- fantastic multimedia, solid multitasking. It had the virtues of the Mac -- a rich GUI that wasn't a bolted-on extra above a crappy old text-mode OS. It (eventually) had the virtues of the ST -- cheap, affordable hardware. It had the virtues of grown-up OSes NT & Unix -- a solid clean OS kernel design, support for a Unix-like shell, C++, standard filesystems, standard networking protocols.

But it was also small, clean and very fast.

What it didn't have were the apps, and it was too early for the Web era and rich Web apps (and possibly Java ones) which might have saved it.

It's mostly forgotten now, which is a shame. It embodied a lot of important lessons.

Some, sadly, are what not to do.

For instance, it switched platforms 3, arguably 4 or 5 times, and that stopped it acquiring any momentum.

It started off with dual AT&T Hobbit CPUs in the prototype BeBox hardware (circa 1993). Be (sensibly) switched the BeBox to dual PowerPC CPUs before release (1995).

It then discontinued its own hardware and ported BeOS to the PowerMac, despite the lack of openly-available documentation for the platform. Jean-Louis Gassée said that the port was despite Apple, rather than aided by them. It also ran on other PREP-compatible PowerPC systems, such as the Power Computing Mac clones.

Around '96-'97, Apple nearly bought Be, but Gassée asked too much. In 1996 Apple went with NeXT instead. (A good move -- if it had bought Be, it'd be long dead, I think. NeXT had far better developer tools, and they got Steve Jobs back into the bargain. Some observers said it was effectively NeXT taking over Apple rather than the reverse.)

Facing actual opposition from Apple, which didn't want other companies selling rival OSes to run on its hardware, and the disappearance of the Mac clones, Be ported the OS again, this time to Pentium-level x86 PCs. This is the version I played around with. My elderly test box in around 1999 -- a single-CPU Cyrix 6x86 at 200 MHz, with IDE disks and about 192MB of RAM -- could cold-boot to the BeOS desktop in about 10 seconds. My fastest Windows machine still can't do that today, off an SSD.

Finally there was BeIA, for "internet appliances", with some very clever compression stuff, booting direct to Opera 4. But in 2000-2001, it was too soon for that to be a viable market -- it was just before widespread broadband or wifi adoption, for instance.

By the time BeOS ran on COTS kit, if you wanted to play around with alternative OSes, there were some fairly decent Linux distros around, with a proper desktop in the form of KDE 1.x. It was vastly bigger, slower, more complicated and more arcane, but in essence, it did everything BeOS did, albeit in a far far slower and clunkier way.

BeOS was fast, elegant, attractively GUI-centric. It was stable, it had superb media support without fancy hardware acceleration -- just the sort of kit Acorn made.

Now, true, it would have looked radically unfamiliar to RISC OS users, but I'm sure they could have built bridges. A RISC OS VM for BeOS, and a port of BBC BASIC, as a minimal starter.

BeOS needed its own great hardware, so instead of running on PowerMacs, it switched to Intel. Acorn had great hardware, but it needed an OS.

It never happened, but perhaps it should have.